Prediction of Clinical Transporter‐Mediated Drug–Drug Interactions via Comeasurement of Pitavastatin and Eltrombopag in Human Hepatocyte Models

A structurally identifiable micro‐rate constant mechanistic model was used to describe the interaction between pitavastatin and eltrombopag, with improved goodness‐of‐fit values through comeasurement of pitavastatin and eltrombopag. Transporter association and dissociation rate constants and passive rates out of the cell were similar between pitavastatin and eltrombopag. Translocation into the cell through transporter‐mediated uptake was six times greater for pitavastatin, leading to pronounced inhibition of pitavastatin uptake by eltrombopag. The passive rate into the cell was 91 times smaller for pitavastatin compared with eltrombopag. A semimechanistic physiologically‐based pharmacokinetic (PBPK) model was developed to evaluate the potential for clinical drug–drug interactions (DDIs). The PBPK model predicted a twofold increase in the pitavastatin peak blood concentration and area under the concentration‐time curve in the presence of eltrombopag in simulated healthy volunteers. The use of structural identifiability supporting experimental design combined with robust micro‐rate constant parameter estimates and a semimechanistic PBPK model gave more informed predictions of transporter‐mediated DDIs.

A structurally identifiable micro-rate constant mechanistic model was used to describe the interaction between pitavastatin and eltrombopag, with improved goodness-of-fit values through comeasurement of pitavastatin and eltrombopag. Transporter association and dissociation rate constants and passive rates out of the cell were similar between pitavastatin and eltrombopag. Translocation into the cell through transporter-mediated uptake was six times greater for pitavastatin, leading to pronounced inhibition of pitavastatin uptake by eltrombopag. The passive rate into the cell was 91 times smaller for pitavastatin compared with eltrombopag. A semimechanistic physiologically-based pharmacokinetic (PBPK) model was developed to evaluate the potential for clinical drug-drug interactions (DDIs). The PBPK model predicted a twofold increase in the pitavastatin peak blood concentration and area under the concentration-time curve in the presence of eltrombopag in simulated healthy volunteers. The use of structural identifiability supporting experimental design combined with robust micro-rate constant parameter estimates and a semimechanistic PBPK model gave more informed predictions of transporter-mediated DDIs.
Pitavastatin, one of the family of 3-hydroxy-3-methyl-glutaryl-CoA (HMG-CoA) reductase inhibitors used to manage hypercholesterolemia, has been determined in vitro to be a substrate of organic anion transporting polypeptide (OATP)1B1 and OATP1B3 (fraction transported of 0.78) and of the efflux transporters breast cancer resistance protein (BCRP) and multidrug resistance associated protein (MRP) 2. 1,2 Pitavastatin is more sensitive to transporter inhibition than rosuvastatin in vitro as well as in healthy volunteers 2 and is therefore a good candidate for evaluating transporter-mediated drugdrug interactions (TrDDIs). Elimination of pitavastatin through metabolism and urinary excretion is relatively small compared with the biliary elimination of pitavastatin (53%). 3 Eltrombopag is a thrombopoietin agonist used in the management of thrombocytopenic purpura, and the dose is individualized based on the platelet count to prevent excessive clotting or a lack of effect. 4 It is highly protein bound, and the adsorption to plasma proteins was included to obtain an inhibition concentration at half of the maximum inhibition (IC 50 ) value that explained the inhibition of rosuvastatin. 5 In vitro studies found eltrombopag to be a substrate of OATP1B1, OATP2B1, organic cation transporter 1 (OCT1),

Study Highlights
WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC? ✔ Currently, most in vitro models are not guided by structural identifiability analysis, relying on substrate-only measurements to evaluate transporter inhibition and subsequent drug-drug interaction (DDI) predictions. WHAT QUESTION DID THIS STUDY ADDRESS? ✔ Whether using micro-rate constants compared with macro-rate constants to describe the comeasurement of uptake of substrate (pitavastatin) and inhibitor (eltrombopag) can improve model fits through structurally identifiable mechanistic models, and the potential for transporter DDIs in the clinic. WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE? ✔ The comeasurement of pitavastatin and eltrombopag guided structurally identifiability analysis and improved model fits through micro-rate constants compared with macro-rate constants in human hepatocytes, with additional information provided regarding transporter binding. The parameter estimates were then scaled to a semimechanistic physiologically-based pharmacokinetics (PBPK) model to predict potential clinical interactions. HOW MIGHT THIS CHANGE DRUG DISCOVERY, DE-VELOPMENT, AND/OR THERAPEUTICS? ✔ Micro-rate constants provide a more dynamic view of binding and translocation, furthering the understanding of transporter pharmacology compared with macrorate constants, which can be used in the development of PBPK models and thereby decrease the risk for clinical transporter-mediated DDIs. and BCRP and is also able to inhibit probe substrates for each transporter. 5,6 The uptake by OATP1B1 is disputed perhaps because of the large amount of nonspecific binding to plastic. 4,6 Structural identifiability analysis considers the uniqueness of the unknown model parameters from the proposed input-output model structure corresponding to the proposed data collection used for parameter estimation. 7-10 This is an important, but often overlooked, theoretical prerequisite to experiment design and parameter estimation because estimates for unidentifiable parameters are effectively meaningless. It is therefore important from a systems pharmacology approach to evaluate, assuming noise-free data, whether the proposed mathematical model is at least structurally locally identifiable. 7-10 Evaluation of TrDDIs in vitro is normally conducted without the comeasurement of both substrate and inhibitor in the same sample, assuming that the inhibitor is equal in the medium and cellular compartments using Michaelis-Menten kinetics. This can lead to the structural unidentifiability of the model, affecting the robustness of estimated parameters, upon which critical decisions may be based. 7 However, unlike the use of Michaelis-Menten kinetics, the structural identifiability of micro-rate constant mechanistic models are unaffected by this assumption. 11,12 Clinical drug-drug interactions (DDIs) have been observed between eltrombopag and rosuvastatin (as a perpetrator 13 ) and lopinavir-ritonavir (as a victim 14 ). The main cause of the interaction of eltrombopag with rosuvastatin in vitro was the result of BCRP inhibition, with minimal inhibition of OATP1B1. 5,15 This was confirmed in a semimechanistic physiologically-based pharmacokinetic (PBPK) model comprising the gastrointestinal tract, liver extracellular space, liver, and a central compartment to adequately describe the interaction between rosuvastatin and eltrombopag. 5 The aim of this article is to describe the TrDDI between pitavastatin (substrate) and eltrombopag (inhibitor) using an in vitro micro-rate constant mechanistic modeling approach. The micro-rate constants were used to describe the dynamic transporter interactions following the comeasurement of both in cryopreserved human hepatocytes. The obtained parameters from the in vitro experiment were then scaled to a semimechanistic PBPK model to evaluate the interaction between pitavastatin and eltrombopag in simulated healthy volunteers and compared with a static model used to predict clinical interactions (R value, US Food and Drug Administration (FDA) guidance document 16 ).

Use of hepatocytes
Human hepatocytes were obtained from BioIVT (Brussels, Belgium; lot number: LYB; 10 donor LiverPool (8 Caucasians, 1 African American, and 1 Hispanic) and thawed according to supplier guidelines in Leibovitz L15 medium. Hepatocytes were kept on ice prior to use and were used within 3 hours of defrosting.

Incubations
Hepatocytes (viability 84-87%) were preincubated in Leibovitz L15 medium at 1.5 × 10 6 cells for 15 minutes at 37ºC with either 0.1% DMSO or 45 nmol/mL eltrombopag. Incubations were started by the addition of pitavastatin (0.3-100 nmol/ mL, 1 × 10 6 cells, 0.35% DMSO, 30 nmol/mL eltrombopag final), and 100 µL samples taken at 0.25-30 minutes. Cells were separated using an oil spin method similar to the literature. 11,17 After separation, the tubes were frozen on dry ice and the bottom layer was cut off. Pitavastatin and eltrombopag were extracted in a stop solution prior to ultra-performance liquid chromatography-high-resolution mass spectrometry analysis (see the Supplemental Material for full experimental and sample extraction methods).

Data analysis
The bottom layer concentrations of pitavastatin and eltrombopag were converted to a cellular concentration using a cellular volume (V cell ) of 3 µL/1 × 10 6 human hepatocytes: 18 where V cell /1 × 10 5 cells was 0.3 µL, and the volume of the bottom layer of cesium chloride was 15 µL. A dilution factor of 100 was used to scale from pmol/mL/1 × 10 5 cells to nmol/mL/1 × 10 6 cells.

Structural identifiability analysis
An efficient computerized method to determine the structural local identifiability uses the IdentifiabilityAnalysis package 9,10 in Wolfram Mathematica 11.3 (Wolfram Research Inc., Champaign, IL). Given a set of ordinary differential equations (ODEs) with an unknown parameter vector and known input and a set of measurable cellular observations (see Supplemental Material), this method gives a Boolean answer to the structural identifiability problem, including the list of any unidentifiable parameters.

Mechanistic modeling in hepatocytes
Transporter kinetics are normally evaluated using Michaelis-Menten kinetics as opposed to through the use of more dynamic micro-rate constants 11 assuming that: 19,20 1. Association of substrate to transporter (k a.X ) is very rapid compared with dissociation (k d.X ) and thus at equilibrium within a short timeframe 2. The amount of substrate at the transporter is much greater than the total amount of transporter (T o ) 3. The translocation of substrate from transporter into the cell (k t.X ) is the rate-limiting step in the transporter-mediated uptake of the substrate If these assumptions hold, then the Michaelis-Menten equation can be derived from the micro-rate constant models (models 1 and 2, Table 1), defined here as macro-rate constant models (models 3 and 4, Table 1 19,20 ). To check the validity of these assumptions, the micro-rate constant and macro-rate constant models with and without the comeasurement of eltrombopag were evaluated both for competitive (models 1 and 3, respectively; Table 1) and noncompetitive inhibition (models 2 and 4, respectively; Table 1).
The final chosen mechanistic model (model 1 with the comeasurement of pitavastatin and eltrombopag; Table 1) was based on the outcome of the structural identifiability analysis and goodness-of-fit statistics (Bayesian information criterion, and the weighted Bayesian information criterion, 21 and the sum of the relative mean square root error (percentage RMSRE); see Supplemental Material). Model 1 was described using a set of ODEs representing the following: the amount of pitavastatin (X = substrate (S) or pitavastatin (P)) and eltrombopag (X = inhibitor (I) or eltrombopag (E)) in the medium (X 1 ), the amount bound to transporter (X 2 ), and the intracellular amount (X 3 ). Passive rate constants for movement into and out of the cell (k f.X and k bX , respectively) as well as uptake transporter-mediated rate constants for association to the transporter (k aX ) and dissociation (k dX ) and translocation (k tX ) from the transporter into the medium and cell, respectively. Metabolism of pitavastatin in the cell was through an elimination rate constant (k e ). The sets of ODEs for all the models (models 1-4; Table 1) tested for structural identifiability and used for parameter estimation are included in the Supplemental Material.

Parameter estimation
Parameter estimation was conducted within Monolix 2018 R2 (Lixoft, Antony, France) for each concentration and experiment as an individual data set, where to ensure positivity, a log-normal distribution was assumed for each parameter, with a proportional residual error model for the observations. Because of the large number of parameters to be estimated in the combined pitavastatin and eltrombopag mechanistic model, the initial estimates for pitavastatin and eltrombopag each were obtained for the micro-rate constant models (no macro-rate constant estimates could be obtained for eltrombopag alone for transporter-mediated uptake).

Clinical TrDDI assessment
In vitro evaluation of the potential for clinical TrDDIs are important in drug development to decrease the risk of adverse events and improve patient quality of life. The FDA in their guidance document for industry 16 described the use of the static area under the concentration-time curve (AUC) difference in the presence and absence of inhibitor (the R value 1 ) used to assess the potential for clinical TrDDIs: where f u.pl is the fraction unbound in the plasma (set to 0.01 for eltrombopag, the minimum value proposed in the FDA draft guidance 16 when plasma protein binding ≥ 0.99), K I is obtained from the micro-rate constant model (through (eltrombopag amount at half of the maximum uptake velocity (V max.up.E ) (K m.up.E )), Eq. 12 in Table 2) and: where peak blood concentration (C max ) is the maximum plasma concentration of the inhibitor (nmol/mL), F a F g is the fraction absorbed multiplied by the intestinal availability (0.5 for eltrombopag 5 ), K a is the absorption rate constant set to 0.0084/min for eltrombopag, 5 Q H is hepatic blood flow ( Table 2), and R b is the blood:plasma ratio of 0.78 for eltrombopag. 5 An R value ≥ 1.1 indicates that a TrDDI is likely. 16 A TrDDI was also assessed using a PBPK model, assuming that only the uptake of pitavastatin is inhibited by eltrombopag in humans (and vice versa), with no inhibition of pitavastatin metabolism (see Figure 1). Pitavastatin is metabolized in vitro by both UDP-glucuronosyltransferase (UGT)1A3 and 2B7 and cytochrome P450 (CYP)2C9 and 1B1 and inhibited by eltrombopag (IC 50 = 2-20 µM, exception CYP1B1). 4,22 However, the in vitro mechanistic models did not support metabolism, and a simulation including the inhibition of metabolism of pitavastatin did not affect the pitavastatin profile.
The micro-rate constants for transporter-mediated uptake and passive uptake were scaled to macro-rate constants (see Table 2) and then empirically scaled up to the human body (83 kg, healthy adult; 23 see Table 2). The model was described using ODEs representing the following: The stomach (X 1 ), the gastrointestinal tract (X 2 ), the liver extracellular space (X 3 ), the liver (X 4 ) with biliary clearance into the gallbladder (X 5 ) and metabolic clearance as well as a central blood or plasma compartment (X 6 ; pitavastatin or eltrombopag, respectively), which was linked to X 3 via Q H . See the Supplemental Material for derived ODEs.
The clinical pharmacokinetic data following an oral 1 mg pitavastatin dose in healthy volunteers 2 were used to validate the pitavastatin PBPK model. As the PBPK model was developed for pitavastatin in the blood, the clinical data were converted to blood (ng/mL) by dividing them by the blood:plasma ratio obtained from the literature (0.425 24 ).
Eltrombopag was assumed only to exist in the plasma, because of the high level of binding of eltrombopag to plasma proteins (99.8% 4 ), and the volume of distribution reported for a semimechanistic PBPK model (2940 mL 5 ). Given the high level of binding, eltrombopag fraction unbound in the plasma values of 0.01, 16 0.002, 4 and 0.005 were considered, with a value of 0.005 giving the best visual fit to the 75 mg clinical data for eltrombopag. 25 The F a F g for eltrombopag was reported as 0.5, 5,26 but was assumed to be 1 in the PBPK model to better visually fit simulations with the clinical eltrombopag data.
The ODE PBPK model (see Figure 1 for schematic and Supplemental Material for the set of ODEs) was developed using desolve 27 in R-studio (v1.1.463, R-Studio Inc., Boston, MA) running R-project version 3.6.0 28 and was similar to that used for eltrombopag and rosuvastatin. 5 All figures in this article were generated either within Inkscape (0.93, Inkscape. org, Boston, MA) or cowplot, ggplot2, 29,30 extrafont, 31 and Cairo. 32

Structural identifiability analysis
The micro-rate constant mechanistic models tested (see Supplemental Material) were at least structurally locally identifiable using the IdentifiabilityAnalysis package, 9,10 with no parameters unknown, as long as the initial amounts of drug and outputs (measurement of cellular pitavastatin with or without measurement of eltrombopag) were known (see Table 1). If the cellular concentrations of pitavastatin and eltrombopag were both measured as outputs, then the macro-rate constant mechanistic models tested were at least structurally locally identifiable with no parameters unknown (see Table 1). If only cellular pitavastatin concentration was measured, the macro-rate constant models were unidentifiable unless one of the parameters was known depending on the inhibition type (see Table 1).
For the PBPK model, assuming known doses of pitavastatin and eltromboag and outputs (measurement of pitavastatin and eltrombopag in the central compartment), the model was at least structurally locally identifiable with no parameters unknown if the following were assumed to be known: Volumes and blood flows and the fraction unbound in the blood or plasma and tissues.

Mechanistic modeling in hepatocytes
The micro-rate constant mechanistic model with comeasurement of pitavastatin and eltrombopag and mutual competitive inhibition (Table 1, model 1), had a smaller total RMSRE compared with the measurement of pitavastatin

Pitavastatin and Eltrombopag PBPK DDI Modeling
Carter et al.
(41% and 47%, respectively; see Table 1). Neither noncompetitive inhibition of pitavastatin uptake nor any of the macro-rate constant models were supported as the best-fitting model (see Table 1) and are not discussed further.
Model 1 (see Table 1 and Figure 1 inset) visually fit the pitavastatin data well (Figure 2, blue lines; Table 1, individual RMSRE = 7%), with an initial increase in the cellular pitavastatin concentration at lower incubation concentrations to a   and k mem = approximate membrane volume (1% of cell = 0.00003 mL)); HPGL, hepatocytes per gram liver = 139; 41 K a.P , first-order absorption rate constant for pitavastatin calculated using the mean residence time following oral (PO) and i.v. (IV) administration (MRT P ); 22 k aX , k dX , k tX , transporter association, dissociation, and translocation rate, respectively; k bile , bile flow rate; k eP , pitavastatin elimination rate constant; k fX and k bX , passive rate constants into and out of the cell; k ge , inverse of the gastric emptying time = 1/10 minutes; 1 K m.up.X , concentration at half V max.up.X ; LWT, liver weight for a 83 kg human male; MW X , molecular weight of pitavastatin (421.46) or eltrombopag (442); NA, not applicable; P dif.X and P def.X , passive diffusion into and out of the cell, respectively; 18 PBPK, physiologically-based pharmacokinetic; T o , total amount of transporters; V cell , volume per 1 × 10 6 cells; V C.X , volume of the central compartment (blood and plasma for pitavastatin and eltrombopag respectively); V ext.H , volume of liver extravascular space; V Gabl , volume of the gall bladder; V L ,·liverspecific gravity = 1695.6 g; 23 V max.up.X , maximum uptake velocity; V med , medium volume per 1 × 10 6 cells = 1 mL; X, pitavastatin (P) or eltrombopag (E).
a Visual fit to t max .
b Visual fit to data.

Pitavastatin and Eltrombopag PBPK DDI Modeling
Carter et al.
maximum followed by a decrease as a result of loss from the cell via metabolism. Pitavastatin uptake was saturated as the incubation concentration increased. In the presence of eltrombopag, the transporter-mediated uptake of pitavastatin was reduced ( Figure 2, red lines), which could be overcome by increasing the pitavastatin incubation concentration until the fits overlapped the pitavastatin-only data (Figure 2, 10-100 nmol/mL). The mechanistic model was able to visually recover the eltrombopag concentrations in the presence of pitavastatin, with an initial decrease to a new minimum (Figure 3). Because of the variability in the measured cellular eltrombopag concentration across the three experiments (Figure 3, points), the data are displayed after normalization to a 30 nmol/mL incubation of eltrombopag alone (Figure 3, "0"). At higher pitavastatin incubation concentrations (30 and 100 nmol/mL), the percentage of the eltrombopag-only control was lower than at lower pitavastatin incubation concentrations. The passive rate constant into the cell for pitavastatin (k fP ) was 380 times lower than the passive rate constant out of the cell (k bP ) and 91 times lower than the passive rate constant into the cell for eltrombopag (k fE ) (see Table 2). The passive rate constant out of the cell for eltrombopag (k bE ) was twofold greater than k bP (see Table 2). Transporter association (k a.X ) Figure 1 Schematic of the developed semimechanistic physiologically-based pharmacokinetic model for the concentrations in the liver compartment (X 4 ) assumed to be involved in the transporter-mediated drug-drug interactions between pitavastatin and eltrombopag, which is linked to the concentration in the central compartment (X 6 ) via the concentration in the liver extracellular space (X 3 ) through Q H . The dose of pitavastatin or eltrombopag are applied as an amount into the stomach (X 1 ), which is then transported into the gastrointestinal (GI) tract (X 2 ) and assumed to be in solution, where X = pitavastatin = P or eltrombopag = E. Drug is absorbed into X 3 where free drug moves into the liver via saturable Michaelis-Menten kinetics. Uptake is inhibited by the opposing drug in X 3 via the respective uptake concentration at half of the maximum uptake velocity (K m.up.X ) shown as the inhibition constant (K I.up ). Passive movement of drug into and out of the liver is through clearances P dif.X and P def.X , respectively. Both drugs are cleared through biliary excretion (CL bi.X ) into the gallbladder (X 5 ), and into X 2 where they can be reabsorbed. Both drugs have metabolic clearance from the liver (CL met.X ) while pitavastatin is also cleared into the urine (CL urine.P ) with the kidney blood flow (Q K ). Inset is the in vitro mechanistic model for pitavastatin and eltrombopag representing medium (S 1 and I 1 ), transporter (S 2 and I 2 ), and within hepatocytes (S 3 and I 3 ) linked through rate constants obtained during parameter estimation (see Supplemental Material for ODEs), which are then scaled accordingly to the whole body (see Eqs. 6-10 in Table 2). k ge , gastric emptying rate; k bile , bile flow rate; K a.X , absorption rate constant of X; Inset parameters: k a.X , association rate constant of X; k d.X , dissociation rate constant of X; k t.X , translocation rate constant of X; T o , total amount of uptake transporter sites; k f.X , passive rate constant for movement of X into the cell from the medium; k b.X , passive rate constant for movement of X into the medium from the cell; k e.P , pitavastatin elimination rate constant.

Pitavastatin and Eltrombopag PBPK DDI Modeling
Carter et al.
and dissociation (k d.X ) rate constants were similar between pitavastatin and eltrombopag (see Table 2). The difference in transport between the two substrates was in the transporter translocation rate constant (k t.X ) where pitavastatin was sixfold greater than eltrombopag (see Table 2).
For pitavastatin, a f u.inc = 0.9 (relative standard error = 5%) and gave an estimated f u.l.P = 0.026 (and Table 2) with an analytical recovery of 55%. For eltrombopag, the f u.inc = 0.04 (relative standard error = 5%) gave an estimated f u.cell = 1.2 × 10 -5 and a low analytical recovery of 10%, making the estimate unreliable. This was reflected in the necessary adjustment of fraction unbound in the liver for eltrombopag in the PBPK model (f u.L.E = 0.4; see Table 2).

PBPK model
Although the PBPK models developed here are based on the drug characteristics and physiologically based simulations of concentration and error, the visual fit for a single 1 mg dose of pitavastatin alone (Figure 4a, blue lines) followed the data reasonably well (Figure 4a,b,  solid points and error bars, extracted from literature 2 ). The C max , time to maximum concentration (t max ), and AUC estimates obtained from the PBPK model for pitavastatin only were similar to the literature 2 (see Table 3). For eltrombopag alone, the mean simulation following a single 75 mg dose (Figure 4c, blue lines) visually fitted the literature plasma concentration 25 reasonably well 25 and closely followed the shape of the plasma concentration vs. time data (Figure 4c,d, points, extracted from literature 25 ). However, the 5% confidence interval (Figure 4c,d, lower dashed lines) increased to large proportions during the eltrombopag elimination phase, likely as a result of a lack of observed elimination of eltrombopag over the hepatocyte uptake time course and a high incubation concentration of eltrombopag. The C max and AUC estimates obtained from the PBPK model following a single 75 mg dose of eltrombopag were similar to the literature, 25 whereas t max was later (see Table 3). Following a 1 mg dose of pitavastatin and a 75 mg dose of eltrombopag, the simulated pitavastatin C max in the blood increased from 23.1 (12.5-37.0) ng/mL to 47.8 (32.9-66.3) ng/mL (Figure 4b, red lines) to above the data for a 2 mg dose of pitavastatin (Figure 4b, open circles, extracted from ref. 22). No effect was seen in the PBPK model fit for eltrombopag in the plasma following a 75 mg dose and a 1 mg dose of pitavastatin (Figure 4d, red lines).
The k aP estimate obtained for pitavastatin in human hepatocytes over 70 seconds in the literature was more than 40-fold greater (7.4 (51%)/nmol/min/1 × 10 6 cells 11 ) than the value obtained here (see Table 2), whereas the k dP and k tP values were of the same order of magnitude (6.3 and 4.3 (85%)/min/1 × 10 6 cells, respectively, 11 and see Table 2). The extended time points taken here up to 30 minutes enabled the errors in k dP to be determined and reduced the error on k tP estimated from the data compared with the literature. 11 If the transporter rate constants are transformed to V max.up and K m.up (Eqs. 6 and 7, respectively, in Table 2), then the difference between pitavastatin and eltrombopag can be clearly seen in the V max.up. (302 (177-639) and 49  pmol/min/1 × 10 6 cells, respectively), and K m.up values (22.1 (16.7-25.5) and 7.1 (7.2-9.6) nmol/mL, respectively). The scaled V max.up.P value was similar to the literature (65-354 pmol/min/1 × 10 6 cells), 11,33,34 whereas K m.up.P was 10-fold lower than the literature (1.4-2 nmol/mL). 11,33,34 The similar values for P dif.P and P def.P confirm experimentally to that assumed in the literature, i.e., passive movement of drug into and out of cell are equal when scaled to a clearance. 17,[33][34][35] However, this assumption only holds when transporter-mediated uptake dominates over passive uptake and metabolism, calculated from the fraction trans- Eltrombopag was reported to show saturable uptake in mouse hepatocytes, attributable to uptake transporters, 5 but only as an inhibitor in the FDA submission document. 26 The transporter-mediated uptake of eltrombopag derived using micro-rate constants, which was not possible using a macro-rate constant model, agrees with the findings from mouse hepatocytes, 5 whereas the F T.E value supports the FDA submission document 26 where passive movement dominates.

TrDDI risk assessment
The clinical risk of TrDDI according to the R value 2 with eltrombopag as the perpetrator and pitavastatin as the victim was below the FDA cut-off (1.07 at 75 mg). The FDA label for eltrombopag states that caution should be taken when concomitantly administering eltrombopag with OATP1B1 substrates. 26 The dose of eltrombopag is closely monitored in the clinic because of its pharmacological effect, 4 and the R values were therefore also calculated with eltrombopag as the victim and pitavastatin as the perpetrator. No effect of pitavastatin on eltrombopag was seen in the experimental data except at a pitavastatin incubation of 30-100 nmol/mL (see Figure 3). This can also be observed with the calculated R value below the cut-offs suggested by the FDA. 16 This is because of the low dose of pitavastatin given (1-4 mg) in the clinic 2,36,37 (with an estimated I in.max of 0.5-3 nmol/mL) compared with the estimated pitavastatin K I of 13 (9.5-14) nmol/mL from the micro-rate constant model.
The large degree of inhibition by eltrombopag on pitavastatin seen in human hepatocytes (Figure 2, red lines) led to the development of a semimechanistic PBPK model (see Figure 1) to evaluate the potential for a clinical TrDDI in a more dynamic environment. The increase in the simulated pitavastatin C max in the presence of eltrombopag from 23.1 (12.5-37.0) ng/mL to 47.8 (32.9-66.5) ng/mL (  calculated R value of 1.06-1.07. The simulated concentration of eltrombopag in the liver extracellular compartment following a 75 mg dose (66,524 ng/mL at 5 minutes) was much larger than the K I.E value calculated from the uptake experiment (3,138 ng/mL), making a TrDDI more likely with pitavastatin as the "victim" drug and eltrombopag as the "perpetrator" drug. The inhibition of rosuvastatin by eltrombopag has been attributed to the inhibition of BCRP with little contribution of inhibition of OATP1B1. 5,15 For the uptake of rosuvastatin to be inhibited to the same extent that pitavastatin was in this study, the calculated IC 50 had to be 10 times more potent. 5 Pitavastatin has been shown to be more sensitive to the inhibition of OATP1B1 compared with rosuvastatin with intravenous rifampicin, 2 and the K I.E estimated here for eltrombopag would therefore also be sufficient clinically to inhibit the uptake of pitavastatin into the liver, leading to a clinical TrDDI. 5,15 The P dif.E value (1.2 × 10 5 mL/min/liver; Table 2) was much greater than the transporter-mediated clearance for eltrombopag (V max.up.E.WB /K m.up.E.WB = 1625 mL/min/liver), making a TrDDI unlikely with eltrombopag as the "victim" drug and pitavastatin as the "perpetrator," seen from the lack of difference in the simulated C max , t max , and AUC values (see Table 3). The same was concluded for eltrombopag and rosuvastatin. 5

CONCLUSIONS
By additionally measuring eltrombopag and relying on the large amount of uptake data for pitavastatin in human hepatocytes in vitro in the presence and absence of eltrombopag, robust uptake kinetics for eltrombopag and pitavastatin were obtained. The use of a mechanistic modeling approach for parameter estimation, guided by structural identifiability to improve experimental design, shows the advantage of co-measurement of both substrate and inhibitor. An approach was presented with structural identifiable micro-rate constant mechanistic models that can aid in promoting systems pharmacology models of transporters and allows the Michaelis-Menten assumptions for transporters to be formally tested. The improvement in the structural identifiability analysis results and in the model fit seen here using micro-rate constants compared with macro-rate constants is in line with previous studies, 11,12 showing the utility of robust micro-rate constant mechanistic models in TrDDI analysis in drug development. The developed semimechanistic PBPK model, based on the inhibition of uptake into the liver only, predicted a likely clinical TrDDI between pitavastatin and eltrombopag when compared with use of the static R value.

Future directions
It is recommended to comeasure substrate and inhibitor concentrations in the same sample during in vitro TrDDI studies, where a comprehensive mechanistic understanding is required. Further work on developing a robust high throughput of the uptake method (see refs. 12,38) across more structurally diverse substrates and inhibitors will help increase confidence on the approach presented here using micro-rate constants. The robust estimation of the parameters for use in the PBPK model, rather than the use of fixed parameters (with the exception of physiological flows and parameters), and the development of a more micro-rate constant-based PBPK model may further the adoption of this approach.
Supporting Information. Supplementary information accompanies this paper on the CPT: Pharmacometrics & Systems Pharmacology website (www.psp-journal.com).